Study On The Extraction Method Of Yellow River Coastline Based On Remote Sensing Image

The middle and lower reaches of the Yellow River are typical wandering rivers,which are characterized by high frequency of river changes,rapid changes of river beaches and islands,and complex shoreline.These determine the complexity of the measurement and extraction of the Yellow River shoreline information.The Yellow River shoreline data can be continuously and dynamically extracted from remote sensing image data,but there are two prominent problems in existing methods.One is that the characteristics of the Yellow River bank line in some remote sensing images are not obvious,which leads to more mixed non-shoreline information in shoreline extraction.The other is that the extraction of multi temporal massive shoreline information takes time and energy,and the extraction efficiency needs to be improved.In view of the above problems,this thesis studies the Yellow River shoreline extraction method based on remote sensing image data.Aiming at the problem of large amount of non-shoreline information mixed in edge detection,the method of extracting the Yellow River shoreline based on wavelet transform is studied.Aiming at the problem of efficient extraction of massive Yellow River shoreline information,a method of the Yellow River shoreline extraction based on deep learning is studied.The main work is as follows:(1)The applicability of classical edge detection operators in the Yellow River shoreline extraction is analyzed.Based on the spectral characteristics of remote sensing images of the Yellow River,the near-infrared band images are selected,and Sobel operator,Prewitt operator,Laplace operator and Canny operator are used to extract the Yellow River shoreline.It is found that under canny operator,the Yellow River coastline extraction has the least missing information and the coastline continuity is the best,which can be used as the preferred method for edge detection after deep learning semantic segmentation.(2)The Yellow River shoreline extraction method based on wavelet transform is studied.In view of the weak filtering ability of the non-shoreline information in the classical edge detection methods,considering that the wavelet transform modulus maxima has the smoothing filtering effect on the non-shoreline information,a method of Yellow River shoreline extraction based on wavelet modulus maximum is proposed,and the experimental results show that the method has good shoreline extraction effect.In order to further filter and de-noising non shoreline information,based on the edge detection of wavelet modulus maximum,a denoising method based on the local timefrequency characteristics of wavelet transform is studied,and the corresponding algorithm and process are designed.The experiment results show that the method can effectively filter most of the non-shoreline information in the Yellow River shoreline extraction.(3)This thesis designs and implements a method of Yellow River shoreline extraction based on deep learning.Aiming at the problem of efficient extraction of massive Yellow River shoreline information,the deep learning based on U-Net is used to extract the Yellow River shoreline automatically,which divides the river course and its nearby features into rivers,farmland,mountains,sandy land and woodland,and then collects and labels the sample data sets of these features based on the image.The lowlevel features and high-level semantic features of the image are extracted through sample training,and the network model is established.Then,the image is segmented based on the training network model.Finally,Canny operator is used to extract the Yellow River shoreline information based on the image segmentation results.In the process of network design,the efficiency of feature extraction is improved by introducing hole convolution,and the accuracy of Yellow River shoreline segmentation is improved by introducing attention mechanism.Corresponding measures are taken to control the over fitting phenomenon and improve the generalization ability of the model.Finally,241 sheets of remote sensing image data are tested and analyzed.Due to the introduction of data augmentation and attention mechanism,the average IOU index of the semantic segmentation model based on deep learning reaches 0.833 on the test set.After post-processing,it can meet the requirements of automatic extraction of Yellow River shoreline from low-resolution remote sensing images,which provides a feasible method for the efficient extraction of Yellow River shoreline.